Three-dimensional image construction techniques have been widely applied in modern researches encompassing various fields. For example, in dental application, the 3D image construction technique is utilized for constructing a 3D dental cast model of a patient as part of the patient's record, which helps dentists to determine the best treatment plan.
Currently, the three-dimensional images can be generated by 3D-model constructing devices such as computerized axial tomography (CT) scanner, magnetic resonance image (MRI) scanner or 3D laser scanner. By stacking a set of CT slices of an object in many parallel planes, we can easily obtain a 3D numerical representation of that object. MRI is based on nuclear magnetic resonance principle (NMR) and uses a gradient magnetic field to align the nuclear magnetization of atoms in an object to acquire raw MRI data (as NMR RF responses), which is transformed into a visual image that represents the internal structure image of the object. The 3D surface laser scanner uses laser light to measure distances from the scanner's sensor to an object in a systematic pattern whereby the 3D surface profile of the object is obtained by calculation of three-dimensional coordinates of the object's surface. Currently, it is widely applied in 3D model industries such as 3D animation, computer game or machine design.
However, small business such as personal studio or academic research laboratory cannot afford the above-mentioned conventional 3D-model constructing devices since they are quite expensive and cost millions of dollars. In addition, some of the devices have safety issues such as the high radiation exposure associated with CT scanner.
Some researches have been conducted to solve the above problems. For example, K. M. Lee et al. disclose a method to reconstruct the three-dimensional image model of an object using shape-from-shading algorithm (“Shape from shading with a linear triangular element surface model” IEEE Trans. Patt. Anal. Mach. Intell., vol. 158, pp. 815-822, 1993). However, when the 3D-image model is reconstructed from only one image captured from one side of an object illuminated by a single light source, the reconstruction accuracy is adversely affected by the shadow cast by the rough surface of the object.